Multi-Model Line: Configure, Load, Measure in 95 s
Automotive parts are large and carry many inspection points, and one line usually rotates through several models, so inspection equipment has to switch with the product. Missing or wrong parts must also be intercepted at the same time as dimensional measurement. This cell turns capture points, inspection items, algorithm parameters, acceptance criteria and calibration parameters into a per-model configuration loaded with one action. Full dimensional measurement and assembly error proofing share one capture, with a 95 second cycle and 0.15 mm correlation to a CMM, subject to site acceptance results.
Multi-model production spends its money on changeover
Dedicated equipment for every workpiece means high investment, large floor space and complex maintenance. One machine forced to handle several models means changing programs, recalibrating and adjusting criteria at every switch, and the downtime can exceed the inspection time itself. The economics of flexible measurement come down to changeover time.
What actually sits inside a configuration file
Capture points decide where the robot stops and which face it looks at. Inspection items decide which dimensions this model needs. Algorithm parameters and acceptance criteria decide what conclusion comes out of the same image. Calibration parameters decide how the coordinate system is built. All five are organized by model and loaded in one action, so a new workpiece needs only a configuration change rather than a new inspection program.
Does changeover need recalibration?
Calibration parameters are stored per model in the configuration and loaded with it. When the mechanical structure and camera mounting are unchanged, recalibration is generally unnecessary and a verification suffices before resuming production. Conversely, once the mechanics have been touched, verification cannot be skipped, and that is where many site problems begin.
Two anomaly types take two paths
Error-proofing anomalies are intercepted immediately, while dimensional out-of-tolerance parts follow the process rule, and the two results map to different actions in the workflow. Merging both into a single reject lane confuses repair and recheck flows. Separated, the intercept mechanism and the verdict work together and the handling path stays clear.
How 95 seconds holds, and how data splits by model
Inspection items load from the configuration per model, the robot path is optimised per model, and measurement and error proofing share capture, avoiding duplicate motion. Those three together keep a complete single-part inspection within 95 seconds. Budget any one of them on an average and the cycle runs away.
Dimensional data enters SPC control charts by model, error-proofing results enter quality records, and both link by workpiece serial number to form a complete quality file per unit. Splitting statistics by model is the only way to see the process capability of a particular product rather than a blended picture.
Configured does not mean free to change
If a new workpiece has a feature layout very different from existing models, points and algorithms still need replanning. Configuration lowers repeated development and debugging cost; it does not remove engineering work. Selection should confirm coverage by product family.
Questions before a multi-model retrofit
What does custom configuration mean exactly?
The inspection program is managed as a configuration file per workpiece model, covering capture points, inspection items, algorithm parameters and criteria. A new workpiece changes the configuration rather than the program.
Must calibration be redone at changeover?
Calibration parameters are stored per model and loaded with the configuration. With mechanics and cameras unchanged, recalibration is usually unnecessary; verification restores production.
How do error proofing and measurement share equipment?
Both share robot capture and calibration: the 3D camera collects dimensional data, and deep learning identifies missing or wrong parts on the same data, so one inspection outputs both result types.
How does inline SPC work?
Each inspection feeds the statistical system in real time, computing process capability by model and dimension and plotting control charts, warning when a trend appears so intervention happens before batch defects.
Is 95 seconds enough for a multi-model line?
The cycle depends on item count and path optimization for each model. 95 seconds is this case's measured level; a new model must be re-verified, subject to site acceptance results.